摘要
针对衣物属性分类的多样性和复杂性,传统算法和并行卷积神经网络难以准确快速地对衣物属性分类,提出了基于卷积神经网络的衣物属性分类方法,从衣物图像不同角度和不同位置特征出发,利用加入了DenseNet网络的模型自动完成特征学习,得到全面的衣物属性分类信息,然后利用cen误差函数优化softmax分类器,提高类间分散性和类内紧密性。结果表明:与并行卷积神经网络和传统算法相比,该网络结构收敛速度更快,在衣物多种属性上分类准确率更高。
For the diversity and complexity of clothing attribute classification,traditional algorithms and parallel convolutional neural networks are difficult to accurately and quickly classify clothing attributes.A clothing attribute classification method based on convolutional neural network is proposed.Starting from the characteristics of clothing images from different angles and different positions,the convolution neural network model with Densenet network structure is added to automatically complete feature learning to obtain comprehensive clothing attribute classification information,and then the softmax classifier is optimized by cen error function to improve inter-class dispersion and intra-class compactness.The experimental results show that,compared with the depthbased learning algorithm and the traditional algorithm,the network structure has faster convergence speed and higher classification accuracy on a variety of clothing attributes.
作者
杨国亮
曾建尤
龚曼
祝靖宇
YANG Guoliang;ZENG Jianyou;GONG Man;ZHU Jingyu(School of Electrical Engineering and Automation,Jiangxi University of Science and Technology,Ganzhou 341000,China)
出处
《重庆理工大学学报(自然科学)》
CAS
北大核心
2020年第1期77-85,共9页
Journal of Chongqing University of Technology:Natural Science
基金
国家自然科学基金资助项目(61763015)
关键词
卷积神经网络
衣物属性分类
特征学习
convolution neural network
classification of clothing attribute
features extraction